Model orders selected using different information criteria for model without triple-wise spike interaction.

Similar to Table 4, the state-space log-linear models () are repeatedly applied to samples of spike sequences with trials. In contrast to Table 4, here a sample of spike sequences is generated from a pairwise () time-dependent log-liner model of neurons. This underlying model is a projection of the full log-liner model examined in Table 4 (dashed lines in Figure 4C) to the pairwise model space. The frequencies of the th order model that minimizes AIC, BIC, and PDIO are counted by repeatedly applying models with different orders () to samples (each with trials). The most frequently selected models are displayed for the different criteria and for different numbers of trials, , as in Table 4. The rows for the KL-risk and MSE display the models that minimize the estimates of the KL-risk and MSE.